xref: /netbsd-src/sys/netinet/tcp_subr.c (revision ce0bb6e8d2e560ecacbe865a848624f94498063b)
1 /*	$NetBSD: tcp_subr.c,v 1.13 1995/04/13 06:36:44 cgd Exp $	*/
2 
3 /*
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *	This product includes software developed by the University of
18  *	California, Berkeley and its contributors.
19  * 4. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)tcp_subr.c	8.1 (Berkeley) 6/10/93
36  */
37 
38 #include <sys/param.h>
39 #include <sys/proc.h>
40 #include <sys/systm.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/socket.h>
44 #include <sys/socketvar.h>
45 #include <sys/protosw.h>
46 #include <sys/errno.h>
47 
48 #include <net/route.h>
49 #include <net/if.h>
50 
51 #include <netinet/in.h>
52 #include <netinet/in_systm.h>
53 #include <netinet/ip.h>
54 #include <netinet/in_pcb.h>
55 #include <netinet/ip_var.h>
56 #include <netinet/ip_icmp.h>
57 #include <netinet/tcp.h>
58 #include <netinet/tcp_fsm.h>
59 #include <netinet/tcp_seq.h>
60 #include <netinet/tcp_timer.h>
61 #include <netinet/tcp_var.h>
62 #include <netinet/tcpip.h>
63 
64 /* patchable/settable parameters for tcp */
65 int 	tcp_mssdflt = TCP_MSS;
66 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
67 int	tcp_do_rfc1323 = 1;
68 
69 extern	struct inpcb *tcp_last_inpcb;
70 
71 /*
72  * Tcp initialization
73  */
74 void
75 tcp_init()
76 {
77 
78 	tcp_iss = 1;		/* wrong */
79 	tcb.inp_next = tcb.inp_prev = &tcb;
80 	if (max_protohdr < sizeof(struct tcpiphdr))
81 		max_protohdr = sizeof(struct tcpiphdr);
82 	if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
83 		panic("tcp_init");
84 }
85 
86 /*
87  * Create template to be used to send tcp packets on a connection.
88  * Call after host entry created, allocates an mbuf and fills
89  * in a skeletal tcp/ip header, minimizing the amount of work
90  * necessary when the connection is used.
91  */
92 struct tcpiphdr *
93 tcp_template(tp)
94 	struct tcpcb *tp;
95 {
96 	register struct inpcb *inp = tp->t_inpcb;
97 	register struct mbuf *m;
98 	register struct tcpiphdr *n;
99 
100 	if ((n = tp->t_template) == 0) {
101 		m = m_get(M_DONTWAIT, MT_HEADER);
102 		if (m == NULL)
103 			return (0);
104 		m->m_len = sizeof (struct tcpiphdr);
105 		n = mtod(m, struct tcpiphdr *);
106 	}
107 	n->ti_next = n->ti_prev = 0;
108 	n->ti_x1 = 0;
109 	n->ti_pr = IPPROTO_TCP;
110 	n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
111 	n->ti_src = inp->inp_laddr;
112 	n->ti_dst = inp->inp_faddr;
113 	n->ti_sport = inp->inp_lport;
114 	n->ti_dport = inp->inp_fport;
115 	n->ti_seq = 0;
116 	n->ti_ack = 0;
117 	n->ti_x2 = 0;
118 	n->ti_off = 5;
119 	n->ti_flags = 0;
120 	n->ti_win = 0;
121 	n->ti_sum = 0;
122 	n->ti_urp = 0;
123 	return (n);
124 }
125 
126 /*
127  * Send a single message to the TCP at address specified by
128  * the given TCP/IP header.  If m == 0, then we make a copy
129  * of the tcpiphdr at ti and send directly to the addressed host.
130  * This is used to force keep alive messages out using the TCP
131  * template for a connection tp->t_template.  If flags are given
132  * then we send a message back to the TCP which originated the
133  * segment ti, and discard the mbuf containing it and any other
134  * attached mbufs.
135  *
136  * In any case the ack and sequence number of the transmitted
137  * segment are as specified by the parameters.
138  */
139 void
140 tcp_respond(tp, ti, m, ack, seq, flags)
141 	struct tcpcb *tp;
142 	register struct tcpiphdr *ti;
143 	register struct mbuf *m;
144 	tcp_seq ack, seq;
145 	int flags;
146 {
147 	register int tlen;
148 	int win = 0;
149 	struct route *ro = 0;
150 
151 	if (tp) {
152 		win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
153 		ro = &tp->t_inpcb->inp_route;
154 	}
155 	if (m == 0) {
156 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
157 		if (m == NULL)
158 			return;
159 #ifdef TCP_COMPAT_42
160 		tlen = 1;
161 #else
162 		tlen = 0;
163 #endif
164 		m->m_data += max_linkhdr;
165 		*mtod(m, struct tcpiphdr *) = *ti;
166 		ti = mtod(m, struct tcpiphdr *);
167 		flags = TH_ACK;
168 	} else {
169 		m_freem(m->m_next);
170 		m->m_next = 0;
171 		m->m_data = (caddr_t)ti;
172 		m->m_len = sizeof (struct tcpiphdr);
173 		tlen = 0;
174 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
175 		xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_int32_t);
176 		xchg(ti->ti_dport, ti->ti_sport, u_int16_t);
177 #undef xchg
178 	}
179 	ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen));
180 	tlen += sizeof (struct tcpiphdr);
181 	m->m_len = tlen;
182 	m->m_pkthdr.len = tlen;
183 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
184 	ti->ti_next = ti->ti_prev = 0;
185 	ti->ti_x1 = 0;
186 	ti->ti_seq = htonl(seq);
187 	ti->ti_ack = htonl(ack);
188 	ti->ti_x2 = 0;
189 	ti->ti_off = sizeof (struct tcphdr) >> 2;
190 	ti->ti_flags = flags;
191 	if (tp)
192 		ti->ti_win = htons((u_int16_t) (win >> tp->rcv_scale));
193 	else
194 		ti->ti_win = htons((u_int16_t)win);
195 	ti->ti_urp = 0;
196 	ti->ti_sum = 0;
197 	ti->ti_sum = in_cksum(m, tlen);
198 	((struct ip *)ti)->ip_len = tlen;
199 	((struct ip *)ti)->ip_ttl = ip_defttl;
200 	(void) ip_output(m, NULL, ro, 0, NULL);
201 }
202 
203 /*
204  * Create a new TCP control block, making an
205  * empty reassembly queue and hooking it to the argument
206  * protocol control block.
207  */
208 struct tcpcb *
209 tcp_newtcpcb(inp)
210 	struct inpcb *inp;
211 {
212 	register struct tcpcb *tp;
213 
214 	tp = malloc(sizeof(*tp), M_PCB, M_NOWAIT);
215 	if (tp == NULL)
216 		return ((struct tcpcb *)0);
217 	bzero((char *) tp, sizeof(struct tcpcb));
218 	tp->seg_next = tp->seg_prev = (struct tcpiphdr *)tp;
219 	tp->t_maxseg = tcp_mssdflt;
220 
221 	tp->t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
222 	tp->t_inpcb = inp;
223 	/*
224 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
225 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
226 	 * reasonable initial retransmit time.
227 	 */
228 	tp->t_srtt = TCPTV_SRTTBASE;
229 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << 2;
230 	tp->t_rttmin = TCPTV_MIN;
231 	TCPT_RANGESET(tp->t_rxtcur,
232 	    ((TCPTV_SRTTBASE >> 2) + (TCPTV_SRTTDFLT << 2)) >> 1,
233 	    TCPTV_MIN, TCPTV_REXMTMAX);
234 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
235 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
236 	inp->inp_ip.ip_ttl = ip_defttl;
237 	inp->inp_ppcb = (caddr_t)tp;
238 	return (tp);
239 }
240 
241 /*
242  * Drop a TCP connection, reporting
243  * the specified error.  If connection is synchronized,
244  * then send a RST to peer.
245  */
246 struct tcpcb *
247 tcp_drop(tp, errno)
248 	register struct tcpcb *tp;
249 	int errno;
250 {
251 	struct socket *so = tp->t_inpcb->inp_socket;
252 
253 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
254 		tp->t_state = TCPS_CLOSED;
255 		(void) tcp_output(tp);
256 		tcpstat.tcps_drops++;
257 	} else
258 		tcpstat.tcps_conndrops++;
259 	if (errno == ETIMEDOUT && tp->t_softerror)
260 		errno = tp->t_softerror;
261 	so->so_error = errno;
262 	return (tcp_close(tp));
263 }
264 
265 /*
266  * Close a TCP control block:
267  *	discard all space held by the tcp
268  *	discard internet protocol block
269  *	wake up any sleepers
270  */
271 struct tcpcb *
272 tcp_close(tp)
273 	register struct tcpcb *tp;
274 {
275 	register struct tcpiphdr *t;
276 	struct inpcb *inp = tp->t_inpcb;
277 	struct socket *so = inp->inp_socket;
278 	register struct mbuf *m;
279 #ifdef RTV_RTT
280 	register struct rtentry *rt;
281 
282 	/*
283 	 * If we sent enough data to get some meaningful characteristics,
284 	 * save them in the routing entry.  'Enough' is arbitrarily
285 	 * defined as the sendpipesize (default 4K) * 16.  This would
286 	 * give us 16 rtt samples assuming we only get one sample per
287 	 * window (the usual case on a long haul net).  16 samples is
288 	 * enough for the srtt filter to converge to within 5% of the correct
289 	 * value; fewer samples and we could save a very bogus rtt.
290 	 *
291 	 * Don't update the default route's characteristics and don't
292 	 * update anything that the user "locked".
293 	 */
294 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
295 	    (rt = inp->inp_route.ro_rt) &&
296 	    ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr != INADDR_ANY) {
297 		register u_long i;
298 
299 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
300 			i = tp->t_srtt *
301 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
302 			if (rt->rt_rmx.rmx_rtt && i)
303 				/*
304 				 * filter this update to half the old & half
305 				 * the new values, converting scale.
306 				 * See route.h and tcp_var.h for a
307 				 * description of the scaling constants.
308 				 */
309 				rt->rt_rmx.rmx_rtt =
310 				    (rt->rt_rmx.rmx_rtt + i) / 2;
311 			else
312 				rt->rt_rmx.rmx_rtt = i;
313 		}
314 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
315 			i = tp->t_rttvar *
316 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
317 			if (rt->rt_rmx.rmx_rttvar && i)
318 				rt->rt_rmx.rmx_rttvar =
319 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
320 			else
321 				rt->rt_rmx.rmx_rttvar = i;
322 		}
323 		/*
324 		 * update the pipelimit (ssthresh) if it has been updated
325 		 * already or if a pipesize was specified & the threshhold
326 		 * got below half the pipesize.  I.e., wait for bad news
327 		 * before we start updating, then update on both good
328 		 * and bad news.
329 		 */
330 		if ((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
331 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh ||
332 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
333 			/*
334 			 * convert the limit from user data bytes to
335 			 * packets then to packet data bytes.
336 			 */
337 			i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
338 			if (i < 2)
339 				i = 2;
340 			i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr));
341 			if (rt->rt_rmx.rmx_ssthresh)
342 				rt->rt_rmx.rmx_ssthresh =
343 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
344 			else
345 				rt->rt_rmx.rmx_ssthresh = i;
346 		}
347 	}
348 #endif /* RTV_RTT */
349 	/* free the reassembly queue, if any */
350 	t = tp->seg_next;
351 	while (t != (struct tcpiphdr *)tp) {
352 		t = (struct tcpiphdr *)t->ti_next;
353 		m = REASS_MBUF((struct tcpiphdr *)t->ti_prev);
354 		remque(t->ti_prev);
355 		m_freem(m);
356 	}
357 	if (tp->t_template)
358 		(void) m_free(dtom(tp->t_template));
359 	free(tp, M_PCB);
360 	inp->inp_ppcb = 0;
361 	soisdisconnected(so);
362 	/* clobber input pcb cache if we're closing the cached connection */
363 	if (inp == tcp_last_inpcb)
364 		tcp_last_inpcb = &tcb;
365 	in_pcbdetach(inp);
366 	tcpstat.tcps_closed++;
367 	return ((struct tcpcb *)0);
368 }
369 
370 void
371 tcp_drain()
372 {
373 
374 }
375 
376 /*
377  * Notify a tcp user of an asynchronous error;
378  * store error as soft error, but wake up user
379  * (for now, won't do anything until can select for soft error).
380  */
381 void
382 tcp_notify(inp, error)
383 	struct inpcb *inp;
384 	int error;
385 {
386 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
387 	register struct socket *so = inp->inp_socket;
388 
389 	/*
390 	 * Ignore some errors if we are hooked up.
391 	 * If connection hasn't completed, has retransmitted several times,
392 	 * and receives a second error, give up now.  This is better
393 	 * than waiting a long time to establish a connection that
394 	 * can never complete.
395 	 */
396 	if (tp->t_state == TCPS_ESTABLISHED &&
397 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
398 	      error == EHOSTDOWN)) {
399 		return;
400 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
401 	    tp->t_rxtshift > 3 && tp->t_softerror)
402 		so->so_error = error;
403 	else
404 		tp->t_softerror = error;
405 	wakeup((caddr_t) &so->so_timeo);
406 	sorwakeup(so);
407 	sowwakeup(so);
408 }
409 
410 void
411 tcp_ctlinput(cmd, sa, ip)
412 	int cmd;
413 	struct sockaddr *sa;
414 	register struct ip *ip;
415 {
416 	register struct tcphdr *th;
417 	extern struct in_addr zeroin_addr;
418 	extern u_char inetctlerrmap[];
419 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
420 
421 	if (cmd == PRC_QUENCH)
422 		notify = tcp_quench;
423 	else if (!PRC_IS_REDIRECT(cmd) &&
424 		 ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0))
425 		return;
426 	if (ip) {
427 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
428 		in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport,
429 			cmd, notify);
430 	} else
431 		in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify);
432 }
433 
434 /*
435  * When a source quench is received, close congestion window
436  * to one segment.  We will gradually open it again as we proceed.
437  */
438 void
439 tcp_quench(inp, errno)
440 	struct inpcb *inp;
441 	int errno;
442 {
443 	struct tcpcb *tp = intotcpcb(inp);
444 
445 	if (tp)
446 		tp->snd_cwnd = tp->t_maxseg;
447 }
448